Air-ground infrared target tracking data set labeling method based on super-pixel structure constraint

Abstract

The invention discloses an air-ground infrared target tracking data set labeling method based on super-pixel structure constraints. The method comprises the following steps: S1, carrying out manual labeling on first M frames of images needing to be processed to obtain an initial training database, carrying out superpixel segmentation on infrared images, realizing external class structure constraint through a clustering algorithm, and extracting positive and negative sample sets of the images; S2, constructing a probability hypergraph model based on a spatial position by using the negative samples extracted in S1, and realizing an internal position structure constraint of the sample set; S3, constructing a class-based probability hypergraph model by using the superpixels of the positive samples extracted in S1, and realizing an internal class structure constraint of the sample set; S4, fusing the saliency maps of the two internal constraints in S2 and S3, and performing effective frame selection and labeling on a target; and S5, updating a training data set and a classifier model through the front M frames of manually labeled images, and training and classifying the images after M frames. The problem of automatically labeling the target in video images by using a visual tracking algorithm is solved.

Description

technical field [0001] The present invention relates to the technical field of image processing, more specifically, it relates to a method for labeling data sets of air-to-ground infrared target tracking based on superpixel structure constraints. Background technique [0002] With the rapid development of big data, cloud computing, machine vision, microelectronics and other technologies, artificial intelligence has been given wings to take off. The use of deep learning for visual processing is currently a relatively successful field of artificial intelligence development. Artificial intelligence has extremely wide applications in target detection, target tracking, scene understanding, image guidance and other fields. The current intelligent algorithm relies heavily on the scale and quality of the dataset, and most of the current dataset production relies on manual labeling. Manual labeling is relatively expensive in terms of labeling time and cost, and cannot meet the needs ...

AI Technical Summary

Problems solved by technology

The current intelligent algorithm relies heavily on the scale and quality of the data set, and most of the current data set production relies on manual labeling. Manual labeling is relatively expensive in terms of labeling time and cost, and cannot meet the needs of fast and accurate labeling of large-scale data sets.

Images